Abstract: Based on the effect of surface roughness on the electromagnetic scattering properties of the radar target, three kinds of novel sub-regionally composite randomly rough surface models with different surface roughnesses are presented to simulate the realistic environment. Instead of the traditional single spectral function model, the rough surface is modeled with different root mean square (RMS) heights and correlation lengths parameters. Finally, the sparse-matrix flat-surface iterative approach (SMFSIA) canonical grid (CAG) method is proposed to analyze the electromagnetic scattering characteristics of the three kinds of surface cases, especially for the influence of different surface roughnesses. Therefore, it provides a theoretical basis for the research, analyzes detection of terrain and landform accords with the actual natural environment.